Clara Bennett: You've been sitting on something all morning, I can tell.
Max Rivera: I have, yeah — okay, picture it: it's a Tuesday, Nvidia's automotive team needs GPU time to test self-driving models, and the person who decides if they get it is... Jensen Huang. Personally. The CEO.
Clara Bennett: Not a procurement process, not an internal pricing mechanism — the CEO.
Max Rivera: Xinzhou Wu — head of automotive at Nvidia — said it on The Verge's Decoder podcast. Almost weekly allocation fights, internal teams competing for the same constrained supply, and when negotiations break down, Huang steps in. Business Insider surfaced it and I — wait, the part that I keep getting stuck on is that Nvidia's own automotive division is in that queue alongside OpenAI and Meta and xAI and Microsoft and Amazon. GPU wait times into mid-2026. Their own people.
Clara Bennett: In practice, that means Wu's team could lose allocation to an external customer Nvidia wants to keep happy. There's no internal priority by default.
Max Rivera: No — and that's what makes this more than a supply story. When your CEO is the tie-breaker for internal chip fights, you're not running a company, you're running a monarchy. That's today's question: is that rational, or is it a structural problem Nvidia just refuses to name?
Clara Bennett: The distinction matters. Rational scarcity management and concentrated authority look identical from the outside.
Max Rivera: Which is exactly why Wu's framing on Decoder is so interesting — he didn't present this as a problem. He presented it like: this is just how it works.
Clara Bennett: That's the thing worth pushing back on, actually — calling it a monarchy assumes the process broke down. Wu's point on Decoder was almost the opposite. Teams negotiate first, weekly, and Huang only enters when they genuinely can't resolve it. That's not collapse. That's a designed escalation path.
Max Rivera: Wait — so the weekly meeting itself is the system working?
Clara Bennett: Exactly that. Think about a restaurant with one working oven and three chefs all need it at the same time. The sous chefs try to sort it out. When they can't agree, the head chef makes the call. Nobody says that restaurant is broken. That's just how scarce, critical resources get allocated. Wu literally said 'sometimes we need Jensen to help' — not 'Jensen takes over.' He steps in at impasse.
Max Rivera: I mean — yeah, okay, that reframes it. But here's what still snags me: in the restaurant, everyone knows the head chef's criteria. At Nvidia, Wu said 'it's all of the above' when asked what drives the trade-offs. That's not a criterion, that's a shrug.
Clara Bennett: That's fair. The opacity is real. But decision rights escalation — and I'll translate that immediately: it just means the harder the call, the higher it goes — that principle doesn't require transparent criteria to be rational. It requires that someone with full strategic context makes the call. Huang has that context. A committee algorithm doesn't.
Max Rivera: The zero trillion dollar business thing. That's the context no algorithm has.
Clara Bennett: Right — and that's why it stays with him. You can't encode 'bet on markets that don't exist yet' into a procurement rule. So the monarchy framing misses that every organization with genuinely scarce, strategically critical resources ends up here. Not as failure. As design.
Max Rivera: So the hot take was wrong. The weekly fight — that's not evidence of a broken company. That IS the company running normally under impossible constraints.
Clara Bennett: Mostly right — but not entirely. Because here's what that clean 'it's a designed escalation' framing quietly papers over: Wu said on Decoder the criteria are 'all of the above.' Near-term revenue, long-term bets. That's not a principle. That's a menu. And if you're an engineer on Nvidia's automotive team who just got your GPU allocation cut thirty percent, you have no way to know if you lost to an OpenAI training run or to a market that literally does not exist yet.
Max Rivera: Wait — the 'zero trillion dollar business' thing. That's the actual crack.
Clara Bennett: That's the crack. Huang's framework explicitly accounts for entirely new markets — not underdeveloped, not emerging — nonexistent. So when the steering committee cuts your Blackwell allocation, you can't audit whether the trade-off was principled because one side of the ledger has no numbers in it.
Max Rivera: So it's not a monarchy — it's, I mean, it's actually subtler than that. It's a process that looks principled from the inside and is completely unverifiable from anywhere else. And that automotive engineer — she asks her manager why the cut happened. Manager says it came from the steering committee. That's it. No further explanation available.
Clara Bennett: And there's no transparency mechanism that would change that. External customers — OpenAI, Meta, Amazon — they can't audit it either. H100 and Blackwell wait times stretch into mid-2026, and the reason you're waiting is, in principle, a bet on a trillion-dollar market that doesn't exist.
Max Rivera: No, I don't buy that that's purely accidental, either. Like — actually, wait — that's a structural incentive Nvidia has to keep scarcity murky. Which is a different thing than scarcity being real.
Clara Bennett: That part we should probably hold — because whether Nvidia has an interest in managing the scarcity narrative, and what Vera Rubin being 'in production' actually means for supply, that's a different layer entirely.
Max Rivera: Right — yeah. But the partial win for the hot take is real: the criteria being 'all of the above' means the process is principled in theory and unauditable in practice. That's not nothing.
Clara Bennett: In practice, that's the whole ballgame. You can have sound decision rights escalation and still have a system no one outside the room can verify. Both things are true simultaneously.
Max Rivera: And that's exactly where the Vera Rubin thing cracks it open for me — because Huang stood up at a Tokyo developer event and said Vera Rubin is already in production, large volumes, dismissing the delay reports. But shipments to major cloud providers aren't expected until the second half of 2026. Those are — wait, those are not the same sentence.
Clara Bennett: Production starting and production meeting demand. Two different claims.
Max Rivera: Right — and if you're a cloud provider still waiting on Vera Rubin in mid-2026, 'it's in production' is technically true and practically useless.
Clara Bennett: The H200-to-China case makes the same point more cleanly. Huang said at Tokyo they've only just begun shipping to China — despite regulatory clearance. Cleared doesn't mean flowing. And Nvidia gets to say both 'we're compliant' and 'supply is constrained' simultaneously. That's not a contradiction. That's a very useful position.
Max Rivera: Huh. So the narrative and the supply reality can diverge — and Nvidia benefits from both versions being partly true.
Clara Bennett: That's the calibrated take. Huang's centralized control is rational — the scarcity is genuine, the 'zero trillion dollar business' bets are real strategic inputs, and GPU wait times stretching into mid-2026 even with Blackwell introduced aren't invented. But Nvidia is simultaneously the supplier, the allocator, and the beneficiary of premium pricing on scarce chips. Those incentives don't vanish because the person at the top is principled.
Max Rivera: Both things are true — and the fact that they're both true is, I mean, that's actually the uncomfortable part. It's not a gotcha. It's a structural tension that never resolves.
Clara Bennett: In practice: you cannot fully separate Huang's legitimate scarcity management from Nvidia's incentive to manage the scarcity narrative. They're load-bearing for the same structure.
Max Rivera: Which means — yeah — the question was never 'is Jensen fair.' The question is whether any single company should be supplier, allocator, and price-setter at 90% market share. That's the thing that doesn't get fixed by a better weekly meeting.
Clara Bennett: And that question — whether any single company should hold 90% market share and be supplier, allocator, and price-setter simultaneously — that's the one that actually gets answered when supply normalizes. Maybe 2027, maybe 2028. If Huang hands the allocation process to something transparent and rules-based when the shortage ends, then scarcity really was the constraint. If the centralized control stays — that's a different answer.
Max Rivera: I mean, that's the actual test, isn't it. Because right now scarcity and principle look identical from the outside. You can't tell them apart. The only way to see which one it actually is, is to wait for the shortage to end and watch what Nvidia does. Does Huang decentralize the allocation keys? Or does it turn out the control was always too useful to release.
Clara Bennett: That's where I landed, yeah. And every lab, every startup, every country trying to build competitive AI at that 90% dependency — they're all waiting on the same answer.
Max Rivera: Not a monarchy with bad vibes. Just one with a really long timeline before you find out. Thanks for thinking through it with me — genuinely.
